Forward Stepwise Algorithm Research Articles

Improving Air Pollution Prediction Modelling Using Wrapper Feature Selection

Feature selection is considered as one of the essential steps in data pre-processing. However, all of the previous studies on predicting PM10 concentration in Malaysia have been limited to statistical method feature selection, and none of these studies used machine-learning approaches. Therefore, the objective of this research is to investigate the influence variables of the PM10 prediction model by using wrapper feature selection to compare the prediction model performance of different wrapper feature selection and to predict the concentration of PM10 for the next day. This research uses 10 years of daily data on pollutant concentrations from two stations (Klang and Shah Alam) obtained from the Department of Environment Malaysia (DOE) from 2009 until 2018. Six wrapper methods (forward selection, backward elimination, stepwise, brute-force, weight-guided and genetic algorithm evolution and the predictive analytics multiple linear regression (MLR) and artificial neural network (ANN)) were implemented in this study. This study found that brute-force is the dominant wrapper method in most of the best models in selecting important features for MLR. Moreover, compared to MLR, ANN provides more advantages regarding model accuracy and permits feature selection in predicting PM10. The overall results revealed that the RMSE value for next day prediction in Klang is 20.728, while the AE value is 15.69. Furthermore, the RMSE value for next day prediction in Shah Alam is 10.004, while the AE value is 7.982. Finally, all of the predicted models in Klang and Shah Alam can be used to predict the PM10 concentrations. This proposed model can be used as a tool for an early warning system in giving air quality information to local authorities in order to formulate air-quality-improvement strategies.

Clinical predictors and outcomes for Legionnaire's disease versus bacteremic pneumococcal pneumonia

BackgroundLegionnaires' disease (LD) is a serious sometimes fatal pneumonia caused by Legionella pneumophila. The clinical manifestations of LD may be similar to those by caused by Streptococcus pneumoniae. As both conditions can be serious illnesses but requiring different antimicrobial therapies, factors that can help differentiate these types of pneumonias can be helpful in the clinical management of hospitalized patients with bacterial pneumonia. This study aimed to compare clinical features and indicators of disease progression in hospitalized patients with community-acquired pneumonia caused by L. pneumophila and bacteremic S. pneumoniae. MethodsWe conducted a retrospective case comparison study of adult patients hospitalized with LD or S. pneumoniae. Data collected included demographic, clinical characteristics, and comorbidities, and outcomes. Data were analyzed using SPS vs 24.0. Multivariable analysis was done using logistic regression with a forward stepwise algorithm. ResultsA total of 106 patients met study criteria. The incidence of LD peaked in summer months and S. pneumoniae peaked in the winter quarter. From multivariable analysis predictors of LD were male gender (OR=21.6, p < 0.001), diarrhea (OR=4.5, p = 0.04), body mass index (BMI) (OR=1.13, p = 0.02), hyponatremia (OR=5.6, p = 0.03 and Charlson weighted index of comorbidity (CWIC) score (OR=0.61, p = 0.01). Patients with S. pneumoniae had higher rates of mechanical ventilation, septic shock, and death than those with LD. ConclusionsOur data suggests that variables that may distinguish LD from S. pneumoniae include male gender, diarrhea, hyponatremia, higher temperature on admission, higher BMI and fewer comorbidities. Bacteremic S. pneumoniae was associated with poorer outcomes than LD including higher rates of septic shock, mechanical ventilation, ICU admission, and death.

Chemometric Study of Fatty Acid Composition of Virgin Olive Oil from Four Widespread Greek Cultivars.

Virgin olive oil (VOO) is one of the key components of the Mediterranean diet owing to the presence of monounsaturated fatty acids and various bioactive compounds. These beneficial traits, which are usually associated with the cultivar genotype, are highlighting the demand of identifying characteristics of olive oil that will ensure its authenticity. In this work, the fatty acid (FA) composition of 199 VOO samples from Koroneiki, Megaritiki, Amfissis, and Manaki cultivars was determined and studied by chemometrics. Olive cultivar greatly influenced the FA composition, namely, oleic acid (from 75.36% for Amfissis to 65.81% for Megaritiki) and linoleic acid (from 13.35% for Manaki to 6.70% for Koroneiki). Spearman’s rho correlation coefficients revealed differences and similarities among the olive oil cultivars. The use of the forward stepwise algorithm identified the FAs arachidonic acid, gadoleic acid, linoleic acid, α-linolenic acid, palmitoleic acid, and palmitic acid as the most significant for the differentiation of samples. The application of linear and quadratic cross-validation discriminant analysis resulted in the correct classification of 100.00% and 99.37% of samples, respectively. The findings demonstrated the special characteristics of the VOO samples derived from the four cultivars and their successful botanical differentiation based on FA composition.

Centennial-scale variability of terrestrial near-surface wind speed over China from reanalysis

AbstractInvestigations of variations and causes of near-surface wind speed (NWS) further understanding of the atmospheric changes and improve the ability of climate analysis and projections. NWS varies on multiple temporal scales; however, the centennial-scale variability in NWS and associated causes over China remains unknown. In this study, we employ the European Centre for Medium-Range Weather Forecasts (ECMWF) twentieth century reanalysis (ERA-20C) to study the centennial-scale changes in NWS from 1900–2010. Meanwhile, a forward stepwise regression algorithm is used to reveal the relationships between NWS and large-scale ocean-atmosphere circulations. The results show three unique periods in annual mean NWS over China from 1900–2010. The annual mean NWS displayed a decreasing trend of -0.87% decade-1 and -11.75% decade-1 from 1900–1925 and 1957–2010, respectively, which were caused by the decreases in the days with strong winds, with trends of -6.64 and -4.66 days decade-1, respectively. The annual mean NWS showed an upward trend of 55.47% decade-1 from 1926–1956, which was caused by increases in the days with moderate (0.43 days decade-1) and strong winds (23.55 days decade-1). The reconstructed wind speeds based on forward stepwise regression algorithm matched well with the original wind speeds; therefore, the decadal changes in NWS over China at centennial-scale were mainly induced by large-scale ocean-atmosphere circulations, with the total explanation power of 66%. The strongest explanation power was found in winter (74%), and the weakest explanation power was found in summer (46%).

Forward stepwise random forest analysis for experimental designs

In experimental designs, it is usually assumed that the data follow normal distributions and the models have linear structures. In practice, experimenters may encounter different types of responses and be uncertain about model structures. If this is the case, traditional methods, such as the ANOVA and regression, are not suitable for data analysis and model selection. We introduce the random forest analysis, which is a powerful machine learning method capable of analyzing numerical and categorical data with complicated model structures. To perform model selection and factor identification with the random forest method, we propose a forward stepwise algorithm and develop Python and R codes based on minimizing the OOB error. Six examples including simulation and case studies are provided. We compare the performance of the proposed method and some frequently used analysis methods. Results show that the forward stepwise random forest analysis, in general, has a high power for identifying active factors and selects models that have high prediction accuracy.

Penerapan Multivariate Adaptive Regression Spline (MARS) Pada Kasus Indeks Pembangunan Manusia di Kawasan Timur Indonesia

Human Development Index (HDI) is an important indicator to measure success in effort to build the quality of human life. Multivariate Adaptive Regression Spline (MARS) is a regression approach developed from the Recursive Partitioning Regression (RPR) method and combined with the spline method to produce a model that is continuous on knots. MARS modeling is determined based on trial and error for a combination of basis function (BF), maximum interaction (MI), and minimum observation (MO) to get the value of minimum smoothing parameter. Automatic determination of knots on MARS uses a forward stepwise and backward stepwise algorithm based on minimum Generalized Cross Validation (GCV) values. The results of this study, obtained a combination value of BF = 52, MI = 3, and MO = 2 with a minimum GCV of 0,00049 then the best MARS model is as follows.Y = –1,045 + 0,446*BF1 – 0,592*BF2 + 0,234*BF3 – 0,352*BF4 + 0,218*BF5 – 0,254*BF6 + 0,338*BF7 – 0,304*BF8 – 3,974*BF9 + 0,097*BF13 – 1,267*BF15 – 41,784*BF16 – 1,845*BF18 – 0,519*BF19 – 0,046*BF21 + 1,275*BF23 – 0,006*BF24 – 0,110*BF25 + 0,021*BF26 + 0,153*BF27 + 14,607*BF28 – 9,276*BF30 – 5,335*BF32 – 0,053*BF34 – 0,219*BF35 – 0,111*BF36 + 0,033*BF37 – 0,238*BF38 + 0,037*BF39 – 0,017*BF41 – 0,039*BF42 – 0,632*BF43 + 0,096*BF45 – 0,147*BF46 + 0,070*BF48 + 0,339*BF49 + 0,017*BF50. The factors that influence HDI based on the level of importance in the best MARS model are per capita expenditure (X9), life expectancy (X7), long school expectations (X8), average school length (X6), poverty severity index (X4), percentage of poor people (X5), percentage of per capita expenditure for food (X12), population density per km2 (X1), and percentage of poor people aged 15 years and over do not work (X11).

The Choice of Induction Regimen Affects Post Remission Survival of Acute Myelogenous Leukemia (AML) Patients with Intermediate or Poor Risk Disease

The fludarabine, high dose cytarabine and G_CSF with or without idarubicin combination regimen, referred to as FLAG+/-Ida, is commonly used as a salvage regimen for relapsed/refractory AML but its use as initial induction therapy has been more limited. The MRC trial 15 compared 2 induction courses of FLAG-Ida regimen to 3+7 (anthracycline plus cytarabine) regimens with etoposide (ADE) and found a higher relapse free survival among patients receiving FLAG-Ida. The findings of the MRC AML 15 trial are not widely applied in the United States as many experts use one course of induction followed by consolidation if patients achieve remission. The impact of choice of induction regimen on post remission survival in non-favorable risk AML patients remains unclear. To address this question, we assessed outcomes of 306 consecutive AML patients between aged 18 to75 years, with non-favorable NCCN risk who were received initial treatment at our center between January of 2009 and July of 2017 with either 3+7 (n=88) or FLAG+/-Ida (n=218). Our center's algorithm is to offer allogeneic HSCT as a consolidative therapy to all AML patients with non-favorable risk in CR1 who are considered transplant candidates. Patient characteristics were as follows: Median age 58(19, 75), male 55%, FLT-3 ITD positive 12%, NCCN risk ( intermediate 49%, poor 51%) and normal cytogenetics 45%. Baseline characteristics were similar between the two treatment groups with FLAG+/-Ida being younger at time of diagnosis (59 vs 66 years, P<0.001). A total of 240 (78%) patients achieved CR 1 after 3+7 (n=67, 76%) or FLAG+/-Ida (n=167, 79%) with no difference in the rate of achieving remission between the two groups. Patients in the FLAG-Ida group were more likely to achieve remission after one course of induction (74% vs 62%, p<0.001) and had a faster time to get into CR (30 days vs 37.5, p<0.001) compared to 3+7. CR was achieved in 84% and 73% of patients with intermediate and poor NCCN risk disease respectively. HSCT was performed in 172 patients accounting for 72% of the 240 patients who achieved CR post induction without any significant difference in transplant rates between FLAG+/-Ida and 3+7. The time from diagnosis to transplant was significantly shorter among CR patients after FLAG+/-Ida compared to 3+7 (FLAG+/- Ida 115 days vs 144 days for 3+7, p<0.001). After a median follow up of 41 months for patients achieving CR, the 3 year post remission OS and DFS was significantly better for patients receiving FLAG-Ida at 54% and 49% compared to 39% and 32% for patients receiving 3+7 respectively( P= 0.01 for both endpoints). In multivariate analysis, we analyzed factors associated with OS and DFS since achieving CR1 including age at CR1 (≤49, 50-64, ≥65), gender, race, NCCN risk, induction regimen, number of cycles to CR1, HSCT (yes, no) and year of CR1. HSCT was modelled as a time dependent covariate. A forward stepwise algorithm was implemented and variables were selected if p value was less than 0.05. Factors associated with r post remission survival included age at CR1 (≤49 vs50-64, HR 0.59, p=.02; ≤49 vs ≥65 HR 0.46, p=0.001), NCCN risk (intermediate vs poor HR 0.54, p<0.001), induction regimen (FlAG+/-Ida vs 3+7 HR 0.62, p=0.01) and receiving HSCT (yes vs no HR 0.68, p=0.02). Factors affecting DFS from time of CR also included age, NCCN risk, induction regimen (FlAG+/-Ida vs 3+7 HR 0.68, p=0.028) and receiving HSCT. This single center analysis shows that among patients with non-favorable risk AML, achieving CR after FLAG+/-Ida has better post remission survival than 3+7. This may be partially explained by the faster time to achieve CR and faster time to HSCT in the FLAG+/- Ida group. DisclosuresSolh:Amgen: Speakers Bureau; Celgene: Speakers Bureau; ADC Therapeutics: Research Funding.

Transplant Rates and Outcomes of Young Acute Myeloid Leukemia (AML) Patients with Intermediate and Poor Risk Disease Referred to a Coordinated Acute Leukemia and Hematopoietic Stem Cell Transplant (HSCT) Program

▪Patients with intermediate and poor risk AML have unsatisfactory outcomes with conventional chemotherapy-based treatment. Allogeneic HSCT offers the best chance of long-term survival for patients who achieve complete remission (CR) and are eligible for such a procedure. There are several potential barriers to receiving an allogeneic HSCT, including but not limited to: achieving CR, timely referral to a transplant center, identifying a donor, and performing the HSCT prior to relapse or significant complications arising from induction and consolidation therapy. To better understand the outcomes of young, newly diagnosed non-favorable risk AML patients, we evaluated 271 consecutive adult patients (age ≤65 years) referred to our center between January of 2009 and July of 2017 for management. Patients were treated in an integrated acute leukemia and transplant management unit and it is our policy to offer intensive chemotherapy to all potentially eligible patients and HSCT to all transplant-eligible, non-favorable risk AML patients that achieve CR, with or without complete blood count recovery. Patient characteristics were as follows: Median age 55 (19, 65) years, male sex 50%, NCCN risk (intermediate 51%, poor 49%), FLT 3 ITD positive 15% and normal cytogenetics in 39%. Remission induction chemotherapy was given to 262patients with FLAG+/-IDA being the most commonly used regimen in 161 patients (61%) followed by 3+7 in 63 (24%)patients. Following induction, 211 (81%) achieved a CR1 after one (n=178), two (n=26) or three (n=7) courses of induction with a median follow up for survivors of 44.5 months. For patients with intermediate and poor risk NCCN categories, CR was achieved in 86% and 73% respectively. HSCT was performed on 146 (69%) patients (NCCN intermediate risk n=83, poor risk n=63): Myeloablative conditioning n=103 (71%), autologous (n=10), matched related donor (n=45), matched unrelated donor (n=51) and haploidentical mismatched related donor (n=40). Of the 65 patients in CR1 who did not receive HSCT, the reasons for not receiving HSCT were as follows: relapse prior to transplant (n=23), lack of psychosocial support (n=14), persistent active infection or organ dysfunction (n=16) at time of transplant evaluation, patient declined transplant (n=7) and poor performance status post induction (n=5). The 3 year estimates for OS and DFS were 50% and 44% respectively. Patients who received HSCT had a better 3 year OS and DFS of 59% and 52% compared to 27% and 27% for patients not receiving HSCT (p<0.001). In multivariate analysis, we analyzed factors associated with OS and DFS among patients in CR1 after induction. The variables considered were age at CR1 (≤49, 50-64, ≥65), gender, race (white, black), NCCN risk, induction regimen, cytogenetics at diagnosis (normal, abnormal), FISH at diagnosis (normal, abnormal), FLT3 ITD status, number of cycles to CR1 (1, ≥2), transplantation and year of CR1 (2008-2011, 2012-2014, 2015-2017). ). Transplantation was modelled as a time-dependent covariate where hazard of mortality for transplantation was only evaluated after transplantation was conducted. A forward stepwise algorithm was implemented and variables were selected if p values were less than 0.05. Factors associated with superior overall survival were age at CR1 (≤49 vs ≥60; HR 0.38, p<0.001), NCCN risk (intermediate vs poor HR 0.51, P<0.001), number of cycles to reach CR1 (1 vs ≥2, HR 0.58, p0.025) and receiving HSCT (Yes vs No, HR 0.71, p0.049). Factors associated with improved DFS were age at CR1 (≤49 vs ≥60 years, HR 0.45, p<0.001), NCCN risk (intermediate vs poor, HR0.53, p0.008) and receiving HSCT (yes vs no, HR 0.57, p0.006). This analysis shows that a timely coordinated approach of remission induction followed by HSCT in the majority of patients with non-favorable risk AML can lead to superior outcomes with more than half of the patients achieving long term survival. DisclosuresSolh:Amgen: Speakers Bureau; Celgene: Speakers Bureau; ADC Therapeutics: Research Funding.

Association Between Acidosis Soon After Reperfusion and Contrast-Induced Nephropathy in Patients With a First-Time ST-Segment Elevation Myocardial Infarction.

BackgroundContrast‐induced nephropathy (CIN) is associated with poor outcomes in patients with acute myocardial infarction. However, the predictors of CIN have yet to be fully elucidated.Methods and ResultsThe study included 273 consecutive patients with a first‐time ST‐segment elevation myocardial infarction who underwent reperfusion within 12 hours of symptom onset. The exclusion criteria were hemodialysis, mechanical ventilation, or previous coronary artery bypass grafting. All patients underwent arterial blood gas analysis soon after reperfusion. CIN was defined as an increase of 0.5 mg/dL in serum creatinine or a 25% increase from baseline between 48 and 72 hours after contrast medium exposure. Acidosis was defined as an arterial blood pH <7.35. CIN was observed in 35 patients (12.8%). Multivariable logistic regression analysis with forward stepwise algorithm revealed a significant association between CIN and the following: reperfusion time, the prevalence of hypertension, peak creatine kinase‐MB, high‐sensitivity C‐reactive protein on admission, and the incidence of acidosis (P<0.05). Multivariable logistic regression analysis revealed that the incidence of acidosis was associated with CIN when adjusted for age, male sex, body mass index, amount of contrast medium used, estimated glomerular filtration rate on admission, glucose level on admission, high‐sensitivity C‐reactive protein on admission, and left ventricular ejection fraction (P<0.05). Moreover, the incidence of acidosis was associated with CIN when adjusted for the Mehran CIN risk score (odds ratio: 2.229, P=0.049).ConclusionsThe incidence of acidosis soon after reperfusion was associated with CIN in patients with a first‐time ST‐segment elevation myocardial infarction.

Predicting Software Abnormal State by using Classification Algorithm

Software aging, also called smooth degradation or chronics, has been observed in a long running software application, accompanied by performance degradation, hang/crash failures or both. The key for software aging problem is how to fast and accurately detect software aging occurrence, which is a hard work due to the long delay before aging appearance. In this paper, two problems about software aging prediction are solved, which are how to accurately find proper running software system variables to represent system state and how to predict software aging state in a running software system with a minor error rate. Firstly, the authors use proposed stepwise forward selection algorithm and stepwise backward selection algorithm to find a proper subset of variables set. Secondly, a classification algorithm is used to model software aging process. Lastly, t-test with k-fold cross validation is used to compare performance of two classification algorithms. In the experiments, the authors find that their proposed method is an efficient way to forecast software aging problems in advance.

Clinical Predictors of Abnormal Esophagogastroduodenoscopy (EGD) Findings in Children

Introduction: To evaluate potential predictors as well as age-specific predictors of abnormal esophagogastroduodenoscopy (EGD) results in children undergoing first EGD. We focused on indications for EGD and documented endoscopic and histologic abnormal findings in different age groups in children aged 0-13 years. Methods: This was a retrospective study of children aged 0-13 years who underwent their first EGD from July 2006 to July 2013 in a pediatric GI center in a regional-referral teaching hospital. Demographics, clinical EGD indications, gross, and histologic EGD findings of the study population were collected. Potential predictors were assessed via univariate and multivariate analysis. Univariate analysis included chi square and Student t test; multivariate analysis was done with logistic regression using a forward stepwise algorithm. Results: During the study period, 749 patients were identified as having had an initial EGD. The 3 most common indications found were failed medical management (51.4%), abdominal pain (45.8%), and vomiting (34.7%). The 4 most common abnormal findings were gastritis (21.9%), reflux esophagitis (12.8%), celiac disease (9.1%) and eosinophilic esophagitis (8.3%). Figure 1 indicates the increase of prevalence of abnormal EGD by age. Three significant predictors of abnormality were found in our study population, which included age (odds ratio [OR] 1.13; p<0.001), black race (OR 2.3; p=0.001), and positive celiac panel (OR 1.9; p=0.007). With age-based analysis, black race remained as a predictor in children <2 years (OR 3.68; p=0.002); bloating/flatulence in children aged 2-7 years (OR 3.6; p=0.014); and female gender (OR 0.49; p=0.01) plus unspecific diarrhea (OR 0.03; p=0.28) in children aged 8-11 years. No factors remained significant for children aged 12 years and older.Figure 1Conclusion: Predictors found in this study for abnormal EGD results in children aged 0-13 years were older age, black race, and had positive celiac panel. We also found that different clinical manifestations have age-based predictive value. Predictors of abnormal EGD results vary by age group. Age-adjusted predictors should be taken into consideration in developing EGD indication algorithm in children.

Incidence of and factors associated with perioperative cardiac arrest within 24 hours of anesthesia for emergency surgery.

PurposeTo determine the incidence of and factors associated with perioperative cardiac arrest within 24 hours of receiving anesthesia for emergency surgery.Patients and methodsThis retrospective cohort study was approved by the ethical committee of Maharaj Nakorn Chiang Mai Hospital, Thailand. We reviewed the data of 44,339 patients receiving anesthesia for emergency surgery during the period from January 1, 2003 to March 31, 2011. The data included patient characteristics, surgical procedures, American Society of Anesthesiologists (ASA) physical status classification, anesthesia information, location of anesthesia performed, and outcomes. Data of patients who had received topical anesthesia or monitoring anesthesia care were excluded. Factors associated with cardiac arrest were identified by univariate analyses. Multiple regressions for the risk ratio (RR) and 95% confidence intervals (CI) were used to determine the strength of factors associated with cardiac arrest. A forward stepwise algorithm was chosen at a P-value <0.05.ResultsThe incidence (within 24 hours) of perioperative cardiac arrest in patients receiving anesthesia for emergency surgery was 163 per 10,000. Factors associated with 24-hour perioperative cardiac arrest in emergency surgery were age of 2 years or younger (RR =1.46, CI =1.03–2.08, P=0.036), ASA physical status classification of 3–4 (RR =5.84, CI =4.20–8.12, P<0.001) and 5–6 (RR =33.98, CI =23.09–49.98, P<0.001), the anatomic site of surgery (upper intra-abdominal, RR =2.67, CI =2.14–3.33, P<0.001; intracranial, RR =1.74, CI =1.35–2.25, P<0.001; intrathoracic, RR =2.35, CI =1.70–3.24, P<0.001; cardiac, RR =3.61, CI =2.60–4.99, P<0.001; and major vascular; RR =3.05, CI =2.22–4.18, P<0.001), respiratory or cardiovascular comorbidities (RR =1.95, CI =1.60–2.38, P<0.001 and RR =1.38, CI =1.11–1.72, P=0.004, respectively), and patients in shock prior to receiving anesthesia (RR =2.62, CI =2.07–3.33, P<0.001).ConclusionThe perioperative incidence of cardiac arrest within 24 hours of anesthesia for emergency surgery was high and associated with multiple factors such as young age (≤2 years old), cardiovascular and respiratory comorbidities, increasing ASA physical status classification, preoperative shock, and surgery site. Perioperative care providers, including surgeons, anesthesiologists, and nurses, should be prepared to manage promptly this high risk group of surgical patients.

Discrimination of mineral waters by electronic tongue, sensory evaluation and chemical analysis

Mineral, spring and tap water samples of different geographical origins (7 classes) were distinguished by various methods, such as sensory evaluation, electronic tongue measurement, inductively coupled plasma atomic emission spectroscopy and ion chromatography. Samples from the same geographical origin were correctly classified by chemical analysis and electronic tongue (100%), but it was found that only 80% classification rate can be achieved by sensory evaluation. Different water brands (different brand names) from the same geographical origin did not show definite differences, as expected. Forward stepwise algorithm selected three chemical parameters namely, chloride (Cl−), sulphate (SO42−) and magnesium (Mg) content and two electronic tongue sensor signals (ZZ and HA) to discriminate according to the geographical origins.

The effect of demographic and clinical factors on the relationship between BMI and PSA levels

Studies have reported lower prostate specific antigen (PSA) levels in men with a higher body mass index (BMI). Additional factors such as diabetes mellitus, benign prostatic hyperplasia (BPH) and certain medications may also affect PSA levels and confound the PSA-BMI association. In this study we evaluated the potential confounding effect of these factors on the obesity-PSA relationship and evaluated the association between these factors and PSA level. The study cohort consisted of 770 population-based controls without a history of prostate cancer (PCa) who participated in a prior PCa study. Demographic, anthropometric, and medical history data were obtained, and PSA level was determined from blood drawn at the time of interview. Linear regression was performed to evaluate the PSA-BMI relationship, adjusting for potential confounders. Finally, a forward stepwise algorithm was used to determine which factors were independently associated with PSA values. With increase in BMI (<25, 25-29, ≥30), the geometric mean PSA level declined (1.18, 1.13, and 0.94, respectively); obese men had a 17% (95% CI 0.70-0.99) lower age-adjusted PSA level compared to normal weight men. However, this relationship was non-significant (P = 0.17) in the multivariate model. Independent predictors of PSA level included age (β = 1.03, 95% CI 1.02-1.04), history of BPH (β = 1.48, 95% CI 1.27-1.72), current statin (β = 0.85, 95% CI 0.74-0.98), and NSAID use (β = 0.84, 95% CI 0.72-0.98). The relationship between obesity and PSA is confounded by a number of factors, which likely explain the observed inverse association previously reported. These results should help in interpreting PSA values in men screened for PCa.

Hypotension after spinal anesthesia for cesarean section: identification of risk factors using an anesthesia information management system

To determine risk factors for developing hypotension after spinal anesthesia for cesarean section to prevent obstetric patients from hypotensive episodes potentially resulting in intrauterine malperfusion and endangering the child. The data from 503 women, having received spinal anesthesia for cesarean sections were investigated using online gathered vital signs and specially checked manual entries employing an anesthesia information management system. Blood pressure, heart rate, and oxygen saturation were measured throughout and hypotension was defined as either a drop in mean arterial blood pressure of >20% from baseline value or readings of <90 mmHg systolic arterial blood pressure. Thirty-two variables were studied for association with hypotensive episodes using univariate analysis and logistic regression employing a forward stepwise algorithm to identify independent variables (P < 0.05). Hypotension was found in 284 cases (56.5%). The univariate analysis identified the neonate's weight, mother's age, body mass index, and peak sensory block height associated with hypotension. Body mass index, age and sensory block height were detected as independent factors for hypotension (odds-ratio: 1.61 each). Knowledge of these risk factors should increase the anesthesiologist's attention to decide for the necessity to employ prophylactic or therapeutic techniques or drugs to prevent the neonate from any risk resulting of hypotension of the mother.

Approximation and learning by greedy algorithms

We consider the problem of approximating a given element $f$ from a Hilbert space $\mathcal{H}$ by means of greedy algorithms and the application of such procedures to the regression problem in statistical learning theory. We improve on the existing theory of convergence rates for both the orthogonal greedy algorithm and the relaxed greedy algorithm, as well as for the forward stepwise projection algorithm. For all these algorithms, we prove convergence results for a variety of function classes and not simply those that are related to the convex hull of the dictionary. We then show how these bounds for convergence rates lead to a new theory for the performance of greedy algorithms in learning. In particular, we build upon the results in [IEEE Trans. Inform. Theory 42 (1996) 2118--2132] to construct learning algorithms based on greedy approximations which are universally consistent and provide provable convergence rates for large classes of functions. The use of greedy algorithms in the context of learning is very appealing since it greatly reduces the computational burden when compared with standard model selection using general dictionaries.

Differing incidences of relevant hypotension with combined spinal-epidural anesthesia and spinal anesthesia.

In this investigation we assessed whether patients receiving spinal anesthesia (SPA) as part of combined spinal-epidural anesthesia (CSE) more often experience relevant hypotension than patients receiving SPA alone. From January 1, 1997, until August 5, 2000, electronic anesthesia records from 1596 patients having received SPA and 1023 patients having received CSE for elective surgery were collected by using a computerized anesthesia record-keeping system. Relevant hypotension was defined as a decrease of mean arterial blood pressure of more than 30% within a 10-min interval and a therapeutic action of the attending anesthesiologist within 20 min after onset. Electronic patient charts were reviewed by using logistic regression with a forward stepwise algorithm to identify independent risk factors that were associated with an increased incidence of hypotension after CSE. Univariate analysis was performed to assess differences in biometric data and relevant risk factors for hypotension between the two procedures. The incidence of relevant hypotension was more frequent with CSE than with SPA alone (10.9% versus 5.0%; P < 0.001). In the multivariate analysis, arterial hypertension (odds ratio, 1.83; 95% confidence interval, 1.21-2.78) and sensory block height >T6 (odds ratio, 2.81; 95% confidence interval, 1.88-4.22) were found to be factors associated with hypotension in the CSE group. Compared with patients receiving SPA alone, patients undergoing CSE had a significantly more frequent prevalence of arterial hypertension and higher sensory block levels (P < 0.01) despite smaller amounts of local anesthetics. In this study, patients receiving CSE had an increased risk for relevant hypotension as compared with patients with SPA alone. Part of this effect seems to be due to the procedure alone and not only because this population is at higher risk. This study, based on a large number of patients with a retrospective design by using on-line recorded data, suggests that spinal anesthesia as part of combined spinal-epidural anesthesia may more often lead to relevant hypotension than spinal anesthesia alone. Preexisting arterial hypertension and a sensory block height exceeding T6 are major risk factors for the development of this complication.

The incidence and risk factors for hypotension after spinal anesthesia induction: an analysis with automated data collection.

We sought to identify factors that are associated with hypotension after the induction of spinal anesthesia (SpA) by using an anesthesia information management system. Hypotension was defined as a decrease of mean arterial blood pressure of more than 30% within a 10-min interval, and relevance was defined as a therapeutic intervention with fluids or pressors within 20 min. From January 1, 1997, to August 5, 2000, data sets from 3315 patients receiving SpA were recorded on-line by using the automatic anesthesia record keeping system NarkoData. Hypotension meeting the predefined criteria occurred in 166 (5.4%) patients. Twenty-nine patient-, surgery-, and anesthesia-related variables were studied by using univariate analysis for a possible association with the occurrence of hypotension after SpA. Logistic regression with a forward stepwise algorithm was performed to identify independent variables (P < 0.05). The discriminative power of the logistic regression model was checked with a receiver operating characteristic curve. Calibration was tested with the Hosmer-Lemeshow goodness-of-fit test. The univariate analysis identified the following variables to be associated with hypotension after SpA: age, weight, height, body mass index, amount of plain bupivacaine 0.5% used for SpA, amount of colloid infusion before puncture, chronic alcohol consumption, ASA physical status, history of hypertension, urgency of surgery, surgical department, sensory block height of anesthesia, and frequency of puncture. In the multivariate analysis, independent factors for relevant hypotension after SpA consisted of three patient-related variables ("chronic alcohol consumption," odds ratio [OR] = 3.05; "history of hypertension," OR = 2.21; and the metric variable "body mass index," OR = 1.08) and two anesthesia-related variables ("sensory block height," OR = 2.32; and "urgency of surgery," OR = 2.84). The area of 0.68 (95% confidence interval, 0.63-0.72) below the receiver operating characteristic curve was significantly greater than 0.5 (P < 0.01). The goodness-of-fit test showed a good calibration of the model (H = 4.3, df = 7, P = 0.7; C = 7.3, df = 8, P = 0.51). This study contributes to the identification of patients with a high risk for hypotension after SpA induction, with the risk increasing two- or threefold with each additional risk factor. By using automated data collection, 5 (chronic alcohol consumption, history of hypertension, body mass index, sensory block height, and urgency of surgery) of 29 variables could be detected as having an association with hypotension after spinal anesthesia induction. The knowledge of these risk factors should be useful in increasing vigilance in those patients most at risk for hypotension, in allowing a more timely therapeutic intervention, or even in suggesting the use of alternative methods of spinal anesthesia, such as titrated continuous or small-dose spinal anesthesia.

Computerize anesthesia record keeping in thoracic surgery--suitability of electronic anesthesia records in evaluating predictors for hypoxemia during one-lung ventilation.

The aim of this retrospective study was to assess the suitability of routine data gathered with a computerized anesthesia record keeping system in investigating predictors for intraoperative hypoxemia (SpO2 < 90%) during one-lung ventilation (OLV) in pulmonary surgery. Over a four-year period data of 705 patients undergoing thoracic surgery (pneumonectomy: 78; lobectomy: 292; minor pulmonary resections: 335) were recorded online using an automated anesthesia record-keeping system. Twenty-six patient-related, surgery-related and anesthesia-related variables were studied for a possible association with the occurrence of intraoperative hypoxemia during OLV. Data were analyzed using univariate and multivariate (logistic regression) analysis (p < 0.05). The model's discriminative power on hypoxemia was checked with a receiver operating characteristic (ROC) curve. Calibration was tested using the Hosmer-Lemeshow goodness-of-fit test. An intraoperative incidence of hypoxemia during OLV was found in 67 patients (9.5%). Using logistic regression with a forward stepwise algorithm, body-mass-index (BMI, p = 0.018) and preoperative existing pneumonia (p = 0.043) could be detected as independent predictors having an influence on the incidence of hypoxemia during OLV. An acceptable goodness-of-fit could be observed using cross validation for the model (C = 8.21, p = 0.370, degrees of freedom, df 8; H = 3.21, p = 0.350, df 3), the discriminative power was poor with an area under the ROC curve of 0.58 [0.51-0.66]. In contrast to conventional performed retrospective studies, data were directly available for analyses without any manual intervention. Due to incomplete information and imprecise definitions of parameters, data of computerized anesthesia records collected in routine are helpful but not satisfactory in evaluating risk factors for hypoxemia during OLV.